Skip to main content

AI Agent 'Elements Claw' Cracks Superconducting Material Discovery

In a breakthrough that feels straight out of a sci-fi lab, researchers have unveiled an AI agent that can independently hunt for new superconducting materials—no human hand-holding required. On July 3, the Alibaba DAMO Academy, together with Renmin University of China and the University of Chinese Academy of Sciences, introduced Elements Claw, the world's first AI agent purpose-built for discovering superconductors.

Traditionally, finding a new superconductor is painfully slow. Scientists rely on trial and error, and even the best databases—like SuperCon—have only logged about 2,000 materials over decades. Elements Claw changes the game entirely. It's built on a "specialist-generalist" architecture, trained on a massive database of 125 million molecules and crystal structures. At its core lies a 10-billion-parameter atomic foundation model called Elements, which can predict a material's superconducting potential with an AUC score of 0.996 and estimate its critical temperature within 1 Kelvin.

Image

What makes Elements Claw truly remarkable is its ability to mimic a human scientist's workflow. It can browse scientific literature, assess whether a material can actually be synthesized, design experiments, and even improve its own algorithms when it stumbles upon something new. In a real-world test, the AI sifted through 2.4 million crystal structures and flagged 68,000 promising superconducting candidates—all in just 28 GPU hours. That's a speed and scale no human team could match.

So far, the team has synthesized and verified four new superconducting materials: HfZrRe4, which the AI designed from scratch, plus Hf21Re25, Zr4VRe7, and Zr3ScRe8, discovered by correcting and reanalyzing existing databases. Their critical temperatures reach up to 6.5K.

Rong Yu, head of Science Intelligence at DAMO Academy, says these results confirm the huge potential of AI agents in materials discovery. To accelerate progress, the team has released the full dataset of 2.4 million stable crystals. Professor Huang Wenbing from Renmin University's Gaoqiang Institute of AI adds that this agent framework could be reused for developing other key materials, like solid-state battery electrolytes, multiphase catalysts, and thermoelectric materials. It's a new chapter for scientific discovery—one where AI isn't just a helper, but a lead researcher.

Key Points

  • Elements Claw is the first AI agent designed specifically for superconducting material discovery.
  • It screened 68,000 candidates from 2.4 million crystal structures in 28 GPU hours.
  • Four new superconductors have been synthesized, including one designed entirely by AI.
  • The underlying atomic model predicts superconducting potential with 0.996 AUC and critical temperature within 1K.
  • The framework is expected to be applied to other materials like battery electrolytes and catalysts.